|
(11) | EP 2 963 949 A1 |
| (12) | EUROPEAN PATENT APPLICATION |
|
|
|
|
|||||||||||||||||||||||
| (54) | Method and apparatus for decoding a compressed HOA representation, and method and apparatus for encoding a compressed HOA representation |
| (57) Encoding of Higher Order Ambisonics (HOA) signals commonly results in high data rates.
A method for low bit-rate encoding frames of an input HOA signal having coefficient
sequences comprises computing (s110) a truncated HOA representation (CT(k)), determining (s111) active coefficient sequences (IC,ACT(k)), estimating (s16) candidate directions (MDIR(k)), dividing (s15) the input HOA signal into a plurality of frequency subbands (f1, ..., fF), estimating (s161) for each of the frequency subbands a subset of candidate directions
(MDIR(k)) as active directions (MDIR(k,f1),..., MDIR(k,fF)) and for each active direction a trajectory, computing (s17) for each frequency
subband directional subband signals from the coefficient sequences of the frequency
subband according to the active directions, calculating (s18) for each frequency subband
a prediction matrix (A(k,f1),...,A(k,fF)) that can be used for predicting the directional subband signals from the coefficient
sequences of the frequency subband using the respective active coefficient sequences
(IC,ACT(k)), and encoding (s19) the candidate directions, active directions, prediction matrices
and truncated HOA representation.
|
[0001] This invention relates to a method for encoding frames of an input HOA signal having a given number of coefficient sequences, a method for decoding a HOA signal, an apparatus for encoding frames of an input HOA signal having a given number of coefficient sequences, and an apparatus for decoding a HOA signal.
Background
[0002] Higher Order Ambisonics (HOA) offers one possibility to represent three-dimensional sound, among other techniques like wave field synthesis (WFS) or channel based approaches like the one known as "22.2". In contrast to channel based methods, a HOA representation offers the advantage of being independent of a specific loudspeaker set-up. This flexibility comes at the expense of a decoding process that is required for the playback of the HOA representation on a particular loudspeaker set-up. Compared to the WFS approach, where the number of required loudspeakers is usually very large, HOA may also be rendered to set-ups consisting of only few loudspeakers. A further advantage of HOA is that the same representation can also be employed without any modification for binaural rendering to head-phones.
[0003] HOA is based on the representation of the so-called spatial density of complex harmonic plane wave amplitudes by a truncated Spherical Harmonics (SH) expansion. Each expansion coefficient is a function of angular frequency, which can be equivalently represented by a time domain function. Hence, without loss of generality, the complete HOA sound field representation actually can be understood as consisting of 0 time domain functions, where 0 denotes the number of expansion coefficients. These time domain functions will be equivalently referred to as HOA coefficient sequences or as HOA channels in the following.
[0004] The spatial resolution of the HOA representation improves with a growing maximum order N of the expansion. Unfortunately, the number of expansion coefficients 0 grows quadratically with the order N, and in particular 0 = (N +1)2. For example, typical HOA representations using order N = 4 require 0 = 25 HOA (expansion) coefficients. According to the above considerations, a total bit rate for the transmission of a HOA representation, given a desired single-channel sampling rate fS and the number of bits Nb per sample, is determined by 0 · fS · Nb. Consequently, transmitting a HOA representation e.g. of order N = 4 with a sampling rate of fS = 48kHz employing Nb = 16 bits per sample results in a bit rate of 19.2MBits/s, which is very high for many practical applications such as e.g. streaming. Thus, a compression of HOA representations is highly desirable. Various approaches for compression of HOA sound field representations were proposed in [4, 5, 6]. These approaches have in common that they perform a sound field analysis and decompose the given HOA representation into a directional and a residual ambient component. The final compressed representation comprises, on the one hand, a number of quantized signals, resulting from the perceptual coding of so called directional and vector-based signals as well as relevant coefficient sequences of the ambient HOA component. On the other hand, it comprises additional side information related to the quantized signals, which is necessary for the reconstruction of the HOA representation from its compressed version.
[0005] A reasonable minimum number of quantized signals for the approaches [4, 5, 6] is eight. Hence, the data rate with one of these methods is typically not lower than 256kbit/s, assuming a data rate of 32kbit/s for each individual perceptual coder. For certain applications, like e.g. audio streaming to mobile devices, this total data rate might be too high. Thus, there is a demand for HOA compression methods addressing distinctly lower data rates, e.g. 128kbit/s.
Summary of the invention
[0006] A new method and apparatus for a low bit-rate compression of Higher Order Ambisonics (HOA) representations of sound fields is disclosed.
[0007] One main aspect of the low-bit rate compression method for HOA representations of sound fields is to decompose the HOA representation into a plurality of frequency sub-bands, and approximate coefficients within each frequency sub-band by a combination of a truncated HOA representation and a representation that is based on a number of predicted directional sub-band signals.
[0008] The truncated HOA representation comprises a small number of selected coefficient sequences, where the selection is allowed to vary over time. E.g. a new selection is made for every frame. The selected coefficient sequences to represent the truncated HOA representation are perceptually coded and are a part of the final compressed HOA representation. In one embodiment, the selected coefficient sequences are de-correlated before perceptual coding, in order to increase the coding efficiency and to reduce the effect of noise unmasking at rendering. A partial de-correlation is achieved by applying a spatial transform to a predefined number of the selected HOA coefficient sequences. For decompression, the de-correlation is reversed by re-correlation. A great advantage of such partial de-correlation is that no extra side information is required to revert the de-correlation at decompression.
[0009] The other component of the approximated HOA representation is represented by a number of directional sub-band signals with corresponding directions. These are coded by a parametric representation that comprises a prediction from the coefficient sequences of the truncated HOA representation. In an embodiment, each directional sub-band signal is predicted (or represented) by a scaled sum of the coefficient sequences of the truncated HOA representation, where the scaling is, in general, complex valued. In order to be able to re-synthesize the HOA representation of the directional sub-band signals for decompression, the compressed representation contains quantized versions of the complex valued prediction scaling factors as well as quantized versions of the directions.
[0010] In one embodiment, a method for encoding (and thereby compressing) frames of an input HOA signal having a given number of coefficient sequences, where each coefficient sequence has an index, comprises steps of
determining a set of indices of active coefficient sequences IC,ACT(k) to be included in a truncated HOA representation,
computing the truncated HOA representation CT(k) having a reduced number of non-zero coefficient sequences (i.e. less non-zero coefficient sequences and thus more zero coefficient sequences than the input HOA signal),
estimating from the input HOA signal a first set of candidate directions MDIR(k),
dividing the input HOA signal into a plurality of frequency subbands, wherein coefficient sequences C̃(k - 1, k, f1,...,F) of the frequency subbands are obtained,
estimating for each of the frequency subbands a second set of directions MDIR(k,f1), ..., MDIR(k,fF), wherein each element of the second set of directions is a tuple of indices with a first and a second index, the second index being an index of an active direction for a current frequency subband and the first index being a trajectory index of the active direction, wherein each active direction is also included in the first set of candidate directions MDIR(k) of the input HOA signal (i.e. active subband directions in the second set of directions are a subset of the first set of full band directions),
for each of the frequency subbands, computing directional subband signals X̃(k - 1, k, f1),..., X̃(k -1,k,fF) from the coefficient sequences C̃(k - 1, k, f1,...,F) of the frequency subband according to the second set of directions MDIR(k,f1),....,MDIR(k,fF) of the respective frequency subband,
for each of the frequency subbands, calculating a prediction matrix A(k,f1),..., A(k,fF) that is adapted for predicting the directional subband signals X̃(k - 1, k, f1,...,F) from the coefficient sequences C̃(k - 1, k, f1,...,F) of the frequency subband using the set of indices of active coefficient sequences IC,ACT(k) of the respective frequency subband, and encoding the first set of candidate directions MDIR(k), the second set of directions MDIR(k,f1),...,MDIR(k,fF), the prediction matrices A(k,f1),...,A(k,fF) and the truncated HOA representation CT(k).
[0011] The second set of directions relates to frequency subbands. The first set of candidate directions relates to the full frequency band. Advantageously, in the step of estimating for each of the frequency subbands the second set of directions, the directions MDIR(k,f1),..., MDIR(k,fF) of a frequency subband need to be searched only among the directions MDIR(k) of the full band HOA signal, since the second set of subband directions is a subset of the first set of full band directions. In one embodiment, the sequential order of the first and second index within each tuple is swapped, ie. the first index is an index of an active direction for a current frequency subband and the second index is a trajectory index of the active direction.
[0012] A complete HOA signal comprises a plurality of coefficient sequences or coefficient channels. A HOA signal in which one or more of these coefficient sequences are set to zero is called a truncated HOA representation herein. Computing or generating a truncated HOA representation comprises generally a selection of coefficient sequences that will or will not be set to zero. This selection can be made according to various criteria, e.g. by selecting as coefficient sequences not to be set to zero those that comprise a maximum energy, or those that are perceptually most relevant, or selecting coefficient sequences arbitrarily etc. Dividing the HOA signal into frequency subbands can be performed by Analysis Filter banks, comprising e.g. Quadrature Mirror Filters (QMF).
[0013] In one embodiment, encoding the truncated HOA representation CT(k) comprises partial decorrelation of the truncated HOA channel sequences, channel assignment for assigning the (correlated or decorrelated) truncated HOA channel sequences y1(k),..., yI(k) to transport channels, performing gain control on each of the transport channels, wherein gain control side information ei(k - 1), βi(k - 1) for each transport channel is generated, encoding the gain controlled truncated HOA channel sequences z1(k),..., zI(k) in a perceptual encoder, encoding the gain control side information ei(k - 1), βi(k - 1), the first set of candidate directions MDIR(k), the second set of directions MDIR(k,f1),..., MDIR(k,fF) and the prediction matrices A(k,f1),...,A(k,fF) in a side information source coder, and multiplexing the outputs of the perceptual encoder and the side information source coder to obtain an encoded HOA signal frame
(k - 1).
[0014] In one embodiment, a computer readable medium has stored thereon executable instructions to cause a computer to perform said method for encoding or compressing frames of an input HOA signal.
[0015] In one embodiment, an apparatus for frame-wise encoding (and thereby compressing) frames of an input HOA signal having a given number of coefficient sequences, where each coefficient sequence has an index comprises a processor and a memory for a software program that when executed on the processor performs steps of the above-described method for encoding or compressing frames of an input HOA signal.
[0016] Further, in one embodiment, a method for decoding (and thereby decompressing) a compressed HOA representation comprises
extracting from the compressed HOA representation a plurality of truncated HOA coefficient sequences ẑ1(k),...,ẑI(k), an assignment vector vAMB,ASSIGN(k) indicating (or containing) sequence indices of said truncated HOA coefficient sequences, subband related direction information MDIR(k+1,f1),...,MDIR(k+1,fF), a plurality of prediction matrices A(k+1,f1),...,A(k+1,fF), and gain control side information e1(k),β1(k),...,eI(k),βI(k), reconstructing a truncated HOA representation ĈT(k) from the plurality of truncated HOA coefficient sequences ẑ1(k), ..., ẑI(k), the gain control side information e1(k),β1(k), ..., eI(k),β1(k) and the assignment vector vAMB,ASSIGN(k),
decomposing in Analysis Filter banks the reconstructed truncated HOA representation ĈT(k) into frequency subband representations
for a plurality of F frequency subbands,
synthesizing in Directional Subband Synthesis blocks for each of the frequency subband representations a predicted directional HOA representation
from the respective frequency subband representation
of the reconstructed truncated HOA representation, the subband related direction information MDIR(k+1,f1),...,MDIR(k+1,fF) and the prediction matrices A(k+1,f1),...,A(k+1,fF),
composing in Subband Composition blocks for each of the F frequency subbands a decoded subband HOA representation
with coefficient sequences
that are either obtained from coefficient sequences of the truncated HOA representation
if the coefficient sequence has an index n that is included in (ie. an element of) the assignment vector vAMB,ASSIGN(k), or otherwise obtained from coefficient sequences of the predicted directional HOA component
provided by one of the Directional Subband Synthesis blocks, and
synthesizing in Synthesis Filter banks the decoded subband HOA representations
to obtain the decoded HOA representation Ĉ(k).
[0017] In one embodiment, the extracting comprises demultiplexing the compressed HOA representation to obtain a perceptually coded portion and an encoded side information portion. In one embodiment, the perceptually coded portion comprises perceptually encoded truncated HOA coefficient sequences
(k), ...,
(k) and the extracting comprises decoding in a perceptual decoder the perceptually encoded truncated HOA coefficient sequences
(k), ...,
(k) to obtain the truncated HOA coefficient sequences ẑ1(k),...,ẑI(k). In one embodiment, the extracting comprises decoding in a side information source decoder the encoded side information portion to obtain the set of subband related directions MDIR(k+1,f1),..., MDIR(k+1,fF), prediction matrices A(k+1,f1),...,A(k+1,fF), gain control side information e1(k),β1(k),...,eI(k),βI(k) and assignment vector vAMB,ASSIGN(k).
[0018] In one embodiment, a computer readable medium has stored thereon executable instructions to cause a computer to perform said method for decoding of directions of dominant directional signals.
[0019] In one embodiment, an apparatus for frame-wise decoding (and thereby decompressing) a compressed HOA representation comprises a processor and a memory for a software program that when executed on the processor performs steps of the above-described method for decoding or decompressing frames of an input HOA signal.
[0020] In one embodiment, an apparatus for decoding a HOA signal comprises
a first module configured to receive indices of a maximum number of directions D for a HOA signal representation to be decoded, a second module configured to reconstruct directions of a maximum number of directions D of the HOA signal representation to be decoded, a third module configured to receive indices of active direction signals per subband, a fourth module configured to reconstruct active direction signals per subband from the reconstructed directions D of the HOA signal representation to be decoded, and a fifth module configured to predict directional signals of subbands, wherein the predicting of a directional signal in a current frame of a subband comprises determining directional signals of a preceding frame of the subband, and wherein a new directional signal is created if the index of the directional signal was zero in the preceding frame and is non-zero in the current frame, a previous directional signal is cancelled if the index of the directional signal was non-zero in the preceding frame and is zero in the current frame, and a direction of a directional signal is moved from a first to a second direction if the index of the directional signal changes from the first to the second direction.
[0021] The subbands are generally obtained from a complex valued filter bank. One purpose of the assignment vector is to indicate sequence indices of coefficient sequences that are transmitted/received, and thus contained in the truncated HOA representation, so as to enable an assignment of these coefficient sequences to the final HOA signal. In other words, the assignment vector indicates, for each of the coefficient sequences of the truncated HOA representation, to which coefficient sequence in the final HOA signal it corresponds. For example, if a truncated HOA representation contains four coefficient sequences and the final HOA signal has nine coefficient sequences, the assignment vector may be [1,2,5,7] (in principle), thereby indicating that the first, second, third and fourth coefficient sequence of the truncated HOA representation are actually the first, second, fifth and seventh coefficient sequence in the final HOA signal.
[0022] Further objects, features and advantages of the invention will become apparent from a consideration of the following description and the appended claims when taken in connection with the accompanying drawings.
Brief description of the drawings
[0023] Exemplary embodiments of the invention are described with reference to the accompanying drawings, which show in
Fig.1 an architecture of a spatial HOA encoder,
Fig.2 an architecture of a direction estimation block,
Fig.3 a perceptual side information source encoder,
Fig.4 a perceptual side information source decoder,
Fig.5 an architecture of a spatial HOA decoder,
Fig.6 a spherical coordinate system,
Fig.7 a direction estimation processing block,
Fig.8 directions, a trajectory index set and coefficients of a truncated HOA representation,
Fig.9 a conventional audio encoder as used in MPEG,
Fig.10 an improved audio encoder as usable in MPEG,
Fig.11 a conventional audio decoder as used in MPEG,
Fig.12 an improved audio decoder as usable in MPEG,
Fig.13 a flow-chart of an encoding method, and
Fig.14 a flow-chart of a decoding method.
Detailed description of preferred embodiments
[0024] One main idea of the proposed low-bit rate compression method for HOA representations of sound fields is to approximate the original HOA representation frame-wise and frequency sub-band-wise, i.e. within individual frequency sub-bands of each HOA frame, by a combination of two portions: a truncated HOA representation and a representation based on a number of predicted directional sub-band signals. A summary of HOA basics is provided further below.
[0025] The first portion of the approximated HOA representation is a truncated HOA version that consists of a small number of selected coefficient sequences, where the selection is allowed to vary over time (e.g. from frame to frame). The selected coefficient sequences to represent the truncated HOA version are then perceptually coded and are a part of the final compressed HOA representation. In order to increase the coding efficiency and to reduce the effect of noise unmasking at rendering, it is advantageous to de-correlate the selected coefficient sequences before perceptual coding. A partial de-correlation is achieved by applying to a predefined number of the selected HOA coefficient sequences a spatial transform, which means the rendering to a given number of virtual loudspeaker signals. A great advantage of that partial de-correlation is that no extra side information is required to revert the de-correlation at decompression.
[0026] The second portion of the approximated HOA representation is represented by a number of directional sub-band signals with corresponding directions. However, these are not conventionally coded. Instead, they are coded as a parametric representation by means of a prediction from the coefficient sequences of the first portion, i.e. the truncated HOA representation. In particular, each directional sub-band signal is predicted by a scaled sum of coefficient sequences of the truncated HOA representation, where the scaling is complex valued in general. Both portions together form a compressed representation of the HOA signal, thus achieving a low bit rate. In order to be able to re-synthesize the HOA representation of the directional sub-band signals for decompression, the compressed representation contains quantized versions of the complex valued prediction scaling factors as well as quantized versions of the directions.
[0027] Particularly important aspects in this context are the computation of the directions and of the complex valued prediction scaling factors, and how to code them efficiently.
Low bit rate HOA compression
[0028] For the proposed low bit rate HOA compression, a low bit rate HOA compressor can be subdivided into a spatial HOA encoding part and a perceptual and source encoding part. An exemplary architecture of the spatial HOA encoding part is illustrated in Fig.1, and an exemplary architecture of a perceptual and source encoding part is depicted in Fig.3. The spatial HOA encoder 10 provides a first compressed HOA representation comprising I signals together with side information that describes how to create a HOA representation thereof. In the Perceptual and Side Information Source Coder 30, these I signals are perceptually encoded in a Perceptual Coder 31, and the side information is subjected to source encoding in a Side Information Source Coder 32. The Side Information Source Coder 32 provides coded side information
Then, the two coded representations provided by the Perceptual Coder 31 and the Side Information Source Coder 32 are multiplexed in a Multiplexer 33 to obtain the low bit rate compressed HOA data stream B.
Spatial HOA encoding
[0029] The spatial HOA encoder illustrated in Fig.1 performs frame-wise processing. Frames are defined as portions of O time-continuous HOA coefficient sequences. E.g. a k-th frame C(k) of the input HOA representation to be encoded is defined with respect to the vector c(t) of time-continuous HOA coefficient sequences (cf. eq. (46)) as
where k denotes the frame index, L denotes the frame length (in samples), O = (N + 1)2 denotes the number of HOA coefficient sequences and TS indicates the sampling period.
Computation of a truncated HOA representation
[0030] As shown in Fig.1, a first step in computing the truncated HOA representation comprises computing 11 from the original HOA frame C(k) a truncated version CT(k). Truncation in this context means the selection of I particular coefficient sequences out of the O coefficient sequences of the input HOA representation, and setting all the other coefficient sequences to zero. Various solutions for the selection of coefficient sequences are known from [4,5,6], e.g. those with maximum power or highest relevance with respect to human perception. The selected coefficient sequences represent the truncated HOA version. A data set
(k) is generated that contains the indices of the selected coefficient sequences. Then, as described further below, the truncated HOA versionCT(k) will be partially de-correlated 12, and the partially de-correlated truncated HOA version CI(k) will be subject to channel assignment 13, where the chosen coefficient sequences are assigned to the available I transport channels. As further described below, these coefficient sequences are then perceptually encoded 30 and are finally a part of the compressed representation. To obtain smooth signals for the perceptual encoding after the channel assignment, coefficient sequences that are selected in the kth frame but not in the (k+1)th frame are determined. Those coefficient sequences that are selected in a frame and will not be selected in the next frame are faded out. Their indices are contained in the data set
(k), which is a subset of
(k). Similarly, coefficient sequences that are selected in the kth frame but were not selected in the (k - 1)th frame are faded in. Their indices are contained in the set
(k), which is also a subset of
(k). For the fading, a window function wOA(l), 1 = 1,..., 2L (such as the one introduced below in eq. (39)) may be used.
[0031] Altogether, if a HOA frame k of the truncated version CT(k) is composed of the L samples of the O individual coefficient sequence frames by
then the truncation can be expressed for coefficient sequence indices n = 1,...,O and sample indices 1 = 1,..., L by
[0032] There are several possibilities for the criteria for the selection of the coefficient sequences. E.g., one advantageous solution is selecting those coefficient sequences that represent most of the signal power. Another advantageous solution is selecting those coefficient sequences that are most relevant with respect to the human perception. In the latter case the relevance may be determined e.g. by rendering differently truncated representations to virtual loudspeaker signals, determining the error between these signals and virtual loudspeaker signals corresponding to the original HOA representation and finally interpreting the relevance of the error, considering sound masking effects.
[0033] A reasonable strategy for selecting the indices in the set
(k) is, in one embodiment, to select always the first OMIN indices 1, ..., OMIN, where OMIN = (NMIN + 1)2 ≤ I and NMIN denotes a given minimum full order of the truncated HOA representation. Then, select the remaining I - OMIN indices from the set {OMIN + 1, ..., OMAX} according to one of the criteria mentioned above, where OMAX = (NMAX + 1)2 ≤ 0 with NMAX denoting a maximum order of the HOA coefficient sequences that are considered for selection. Note that OMAX is the maximum number of transferable coefficients per sample, which is less than or equal to the total number O of coefficients. According to this strategy, the truncation processing block 11 also provides a so-called assignment vector
whose elements vA,i(k), i = 1, ..., I - OMIN, are set according to
where n (with n ≥ OMIN + 1) denotes the HOA coefficient sequence index of the additionally selected HOA coefficient sequence of C(k) that will later be assigned to the i-th transport signal yi(k). The definition of yi(k) is given in eq.(10) below. Thus, the first OMIN rows of CT(k) comprise by default the HOA coefficient sequences 1,..., OMIN, and among the following O - OMIN (or OMAX - OMIN, if O = OMAX) rows of CT(k), there are I-OMIN rows that comprise frame-wise varying HOA coefficient sequences whose indices are stored in the assignment vector vA(k). Finally, the remaining rows of CT(k) comprise zeroes. Consequently, as will be described below, the first (or last, as in eq.(10)) OMIN of the available I transport signals are assigned by default to HOA coefficient sequences 1,..., OMIN, and the remaining I-OMIN transport signals are assigned to frame-wise varying HOA coefficient sequences whose indices are stored in the assignment vector vA(k).
Partial de-correlation
[0034] In the second step, a partial de-correlation 12 of the selected HOA coefficient sequences is carried out in order to increase the efficiency of the subsequent perceptual encoding, and to avoid coding noise unmasking that would occur after matrixing the selected HOA coefficient sequences at rendering. An exemplary partial de-correlation 12 is achieved by applying a spatial transform to the first OMIN selected HOA coefficient sequences, which means the rendering to OMIN virtual loudspeaker signals. The respective virtual loudspeaker positions are expressed by means of a spherical coordinate system shown in Fig.6, where each position is assumed to lie on the unit sphere, i.e. to have a radius of 1. Hence, the positions can be equivalently expressed by directions Ωj = (θj, φj) with 1 ≤ j ≤ OMIN, where θj and φj denote the inclinations and azimuths, respectively (see further below for the definition of the spherical coordinate system). These directions should be distributed on the unit sphere as uniformly as possible (see e.g. [2] on the computation of specific directions). Note that, since HOA in general defines directions in dependence of NMIN, actually
is meant whereby Ωj is written herein.
[0035] In the following, the frame of all virtual loudspeaker signals is denoted by
where wj(k) denotes the k-th frame of the j-th virtual loudspeaker signal. Further, ΨMIN denotes the mode matrix with respect to the virtual directions Ωj, with 1 ≤ j ≤ OMIN. The mode matrix is defined by
with
indicating the mode vector with respect to the virtual direction Ωi. Each of its elements
denotes the real valued Spherical Harmonics function defined below (see eq.(48)). Using this notation, the rendering process can be formulated by the matrix multiplication
[0036] The signals of the intermediate representation CI(k), which is output of the partial de-correlation 12, are hence given by
Channel assignment
[0037] After having computed the frame of the intermediate representation CI(k), its individual signals cI,n(k) with n ∈
(k) are assigned 13 to the available I channels, to provide the transport signals yi(k), i = 1, ..., I, for perceptual encoding. One purpose of the assignment 13 is to avoid discontinuities of the signals to be perceptually encoded, which might occur in a case where the selection changes between successive frames. The assignment can be expressed by
Gain control
[0038] Each of the transport signals yi(k) is finally processed by a Gain Control unit 14, where the signal gain is smoothly modified to achieve a value range that is suitable for the perceptual encoders. The gain modification requires a kind of look-ahead in order to avoid severe gain changes between successive blocks, and hence introduces a delay of one frame. For each transport signal frame yi(k), the Gain Control units 14 either receive or generate a delayed frame yi(k - 1), i = 1, ..., I. The modified signal frames after the gain control are denoted by zi(k - 1), i = 1,...,I. Further, in order to be able to revert in a spatial decoder any modifications made, gain control side information is provided. The gain control side information comprises the exponents ei(k - 1) and the exception flags βi(k -1), i = 1,...,I. A more detailed description of the Gain Control is available e.g. in [9], Sect.C.5.2.5, or [3]. Thus, the truncated HOA version 19 comprises gain controlled signal frames zi(k - 1) and gain control side information ei(k - 1), βi(k - 1), i = 1,...,I.
Analysis Filter Banks
[0039] As mentioned above, the approximated HOA representation is composed of two portions, namely the truncated HOA version 19 and a component that is represented by directional sub-band signals with corresponding directions, which are predicted from the coefficient sequences of the truncated HOA representation. Hence, to compute a parametric representation of the second portion, each frame of an individual coefficient sequence of the original HOA representation cn(k), n = 1, ..., O, is first decomposed into frames of individual sub-band signals c̃n(k, f1), ..., c̃n(k, fF). This is done in one or more Analysis Filter Banks 15. For each sub-band fj, j = 1, ..., F, the frames of the sub-band signals of the individual HOA coefficient sequences may be collected into the sub-band HOA representation
[0040] The Analysis Filter Banks 15 provide the sub-band HOA representations to a Direction Estimation Processing block 16 and to one or more computation blocks 17 for directional sub-band signal computation.
[0041] In principle, any type of filters (i.e. any complex valued filter bank, e.g. QMF, FFT) may be used in the Analysis Filter Banks 15. It is not required that a successive application of an analysis and a corresponding synthesis filter bank provides the delayed identity, which would be what is known as perfect reconstruction property. Note that, in contrast to the HOA coefficient sequences cn(k), their sub-band representations c̃n(k, fj) are generally complex valued. Further, the sub-band signals c̃n(k, fj) are in general decimated in time, compared to the original time-domain signals. As a consequence, the number of samples in the frames c̃n(k, fj) is usually distinctly smaller than the number of samples in the time-domain signal frames cn(k), which is L.
[0042] In one embodiment, two or more sub-band signals are combined into sub-band signal groups, in order to better adapt the processing to the properties of the human hearing system. The bandwidths of each group can be adapted e.g. to the well-known Bark scale by the number of its sub-band signals. That is, especially in the higher frequencies two or more groups can be combined into one. Note that in this case each sub-band group consists of a set of HOA coefficient sequences
where the number of extracted parameters is the same as for a single sub-band. In one embodiment, the grouping is performed in one or more sub-band signal grouping units (not explicitly shown), which may be incorporated in the Analysis Filter Bank block 15.
Direction Estimation
[0043] The Direction Estimation Processing block 16 analyzes the input HOA representation and computes for each frequency sub-band fj, j = 1, ..., F, a set
(k,fj) of directions of sub-band general plane wave functions that add a major contribution to the sound field. In this context, the term "major contribution" may for instance refer to the signal power being higher as the signal power of sub-band general plane waves impinging from other directions. It may also refer to a high relevance in terms of the human perception. Note that, where sub-band grouping is used, instead of a single sub-band also a sub-band group can be used for the computation of
(k,fj).
[0044] During decompression, artifacts in the predicted directional sub-band signals might occur due to changes of the estimated directions and prediction coefficients between successive frames. In order to avoid such artifacts, the direction estimation and prediction of directional sub-band signals during encoding are performed on concatenated long frames. A concatenated long frame consists of a current frame and its predecessor. For decompression, the quantities estimated on these long frames are then used to perform overlap add processing with the predicted directional sub-band signals.
[0045] A straight forward approach for the direction estimation would be to treat each sub-band separately. For the direction search, in one embodiment, e.g. the technique proposed in [7] may be applied. This approach provides, for each individual sub-band, smooth temporal trajectories of direction estimates, and is able to capture abrupt direction changes or onsets. However, there are two disadvantages with this known approach. First, the independent direction estimation in each sub-band may lead to the undesired effect that, in the presence of a full-band general plane wave (e.g. a transient drum beat from a certain direction), estimation errors in the individual sub-directions may lead to sub-band general plane waves from different directions that do not add up to the desired full-band version from one single direction. In particular, transient signals from certain directions are blurred.
[0046] Second, considering the intention to obtain a low bit-rate compression, the total bit-rate resulting from the side information must be kept in mind. In the following, an example will show that the bit rate for such naive approach is rather high. Exemplarily, the number of sub-bands F is assumed to be 10, and the number of directions for each sub-band (which corresponds to the number of elements in each set
(k, fj)) is assumed to be 4. Further, it is assumed to perform for each sub-band the search on a grid of Q = 900 potential direction candidates, as proposed in [9]. This requires ┌log2(Q)┐ = 10 bits for the simple coding of a single direction. Assuming a frame rate of about 50 frames per second, a resulting overall data rate is
just for a coded representation of the directions. Even if a frame rate of 25 frames per second is assumed, the resulting data rate of 10 kbit/s is still rather high.
[0047] As an improvement, the following method for direction estimation is used in a Direction Estimation block 20, in one embodiment. The general idea is illustrated in Fig.2.
[0048] In a first step, a Full-band Direction Estimation block 21 performs a preliminary full-band direction estimation, or search, on a direction grid that consists of Q test directions ΩTEST,q , q = 1, ..., Q, using the concatenated long frame
where C(k) and C(k - 1) are the current and previous input frames of the full-band original HOA representation. This direction search provides a number of D(k) ≤ D direction candidates ΩCAND,d(k), d = 1,..., D(k), which are contained in the set
(k), i.e.
[0049] A typical value for the maximum number of direction candidates per frame is D = 16. The direction estimation can be accomplished e.g. by the method proposed in [7]: the idea is to combine the information obtained from a directional power distribution of the input HOA representation with a simple source movement model for the Bayesian inference of the directions.
[0050] In a second step, a direction search is carried out for each individual sub-band by a Sub-band Direction Estimation block 22 per sub-band (or sub-band group). However, this direction search for sub-bands needs not consider the initial full direction grid consisting of Q test directions, but rather only the candidate set
(k), comprising only D(k) directions for each sub-band. The number of directions for the fj-th sub-band, j = 1,..., F, denoted by DSB(k, fj), is not greater than DSB, which is typically distinctly smaller than D, e.g. DSB = 4. Like the full-band direction search, the sub-band related direction search is also performed on long concatenated frames of sub-band signals
consisting of the previous and current frame. In principle, the same Bayesian inference methods as for the full-band related direction search may be applied for the sub-band related direction search.
[0051] The direction of a particular sound source may (but needs not) change over time. A temporal sequence of directions of a particular sound source is called "trajectory" herein. Each subband related direction, or trajectory respectively, gets an unambiguous index, which prevents mixing up different trajectories and provides continuous directional sub-band signals. This is important for the below-described prediction of directional sub-band signals. In particular, it allows exploiting temporal dependencies between successive prediction coefficient matrices A(k, fj) defined further below. Therefore, the direction estimation for the fj-th sub-band provides the set
(k, fj) of tuples. Each tuple consists of, on the one hand, the index d ∈
(k, fj) ⊆ {1, ..., DSB} identifying an individual (active) direction trajectory, and on the other hand, the respective estimated direction ΩSB,d(k, fj), i.e.
[0052] By definition, the set {ΩSB,d(k, fj)|d ∈
(k, fj)} is a subset of
(k) for each j = 1, ..., F, since the sub-band direction search is performed only among the current frame's direction candidates ΩCAND,d(k), d = 1, ..., D(k), as mentioned above. This allows a more efficient coding of the side information with respect to the directions, since each index defines one direction out of D(k) instead of Q candidate directions, with D(k) ≤ Q. The index d is used for tracking directions in a subsequent frame for creating a trajectory. As shown in Fig.2 and described above, a Direction Estimation Processing block 16 in one embodiment comprises a Direction Estimation block 20 having a Full-band Direction Estimation block 21 and, for each sub-band or sub-band group, a Sub-band Direction Estimation block 22. It may further comprise a Long Frame Generating block 23 that provides the above-mentioned long frames to the Direction Estimation block 20, as shown in Fig.7. The Long Frame Generating block 23 generates long frames from two successive input frames having a length of L samples each, using e.g. one or more memories. Long frames are herein indicated by "―" and by having two indices, k-1 and k. In other embodiments, the Long Frame Generating block 23 may also be a separate block in the encoder shown in Fig.1, or incorporated in other blocks.
Computation of directional sub-band signals
[0053] Returning to Fig.1, sub-band HOA representation frames
provided by the Analysis Filter Bank 15 are also input to one or more Directional Sub-band Signal Computation blocks 17. In the Directional Sub-band Signal Computation blocks 17, the long frames of all DSB potential directional sub-band signals x̃d(k - 1; k; fj), d = 1, ..., DSB, are arranged in a matrix x̃(k - 1; k; fj) as
[0054] Further, the frames of the inactive directional sub-band signals, i.e. those long signal frames x̃d(k - 1; k; fj) whose index d is not contained within the set
(k, fj), are set to zero.
[0055] The remaining long signal frames x̃d(k - 1; k; fj), i.e. those with index d ∈
(k, fj), are collected within the matrix
One possibility to compute the active directional sub-band signals contained therein is to minimize the error between their HOA representation and the original input sub-band HOA representation. The solution is given by
where (·)+ denotes the Moore-Penrose pseudo-inverse and
denotes the mode matrix with respect to the direction estimates in the set {ΩSB,d(k, fj)|d ∈
(k, fj)}. Note that in the case of sub-band groups a set of directional sub-band signals x̃ACT(k - 1; k; fj) is computed from the multiplication of one matrix (ΨSB(k, fj))+ by all HOA representations
of the group. Note that long frames can be generated by one or more further Long Frame Generating blocks, similar to the one described above. Similarly, long frame can be decomposed into frames of normal length in Long Frame Decomposition blocks. In one embodiment, the blocks 17 for the computation of directional sub-bands provide on their outputs long frames x̃ACT(k - 1; k; fj), j = 1, ..., F, towards the Directional Sub-band Prediction blocks 18.
Prediction of directional sub-band signals
[0056] As mentioned above, the approximate HOA representation is partly represented by the active directional sub-band signals, which, however, are not conventionally coded. Instead, in the presently described embodiments a parametric representation is used in order to keep the total data rate for the transmission of the coded representation low. In the parametric representation, each active directional sub-band signal x̃d(k - 1; k; fj), i.e. with index d ∈
(k, fj), is predicted by a weighted sum of the coefficient sequences of the truncated sub-band HOA representation c̃n(k - 1, fj) and c̃n(k, fj), where n ∈
(k - 1) and where the weights are complex valued in general.
[0057] Hence, assuming x̃P(k - 1; k; fj) to represent the predicted version of x̃(k - 1; k; fj), the prediction is expressed by a matrix multiplication as
where
is the matrix with all weighting factors (or, equivalently, prediction coefficients) for the sub-band fj. The computation of the prediction matrices A(k, fj) is performed in one or more Directional Sub-band Prediction blocks 18. In one embodiment, one Directional Sub-band Prediction block 18 per sub-band is used, as shown in Fig.1. In another embodiment, a single Directional Sub-band Prediction block 18 is used for multiple or all sub-bands. In the case of sub-band groups, one matrix A(k, fj) is computed for each group; however, it is multiplied by each HOA representations
of the group individually, creating a set of matrices x̃P(k - 1; k; fj) per group. Note that per construction all rows of A(k, fj) except for those with index d ∈
(k, fj) are zero. This means that only the active directional sub-band signals are predicted. Further, all columns of A(k, fj) except for those with index n ∈
(k - 1) are also zero. This means that, for the prediction, only those HOA coefficient sequences are considered that are transmitted and available for prediction during HOA decompression.
[0058] The following aspects have to be considered for the computation of the prediction matrices A(k, fj).
[0059] First, the original truncated sub-band HOA representation
will generally not be available at the HOA decompression. Instead, a perceptually decoded version
of it will be available and used for the prediction of the directional sub-band signals.
[0060] At low bit rates, typical audio codecs (like AAC or USAC) use spectral band replication (SBR), where the lower and mid frequencies of the spectrum are conventionally coded, while the higher frequency content (starting e.g. at 5kHz) is replicated from the lower and mid frequencies using extra side information about the high-frequency envelope.
[0061] For that reason, the magnitude of the reconstructed sub-band coefficient sequences of the truncated HOA component
after perceptual decoding resembles that of the original one,
However, this is not the case for the phase. Hence, for the high frequency sub-bands it does not make sense to exploit any phase relationships for the prediction by using complex valued prediction coefficients. Instead, it is more reasonable to use only real valued prediction coefficients. In particular, defining the index jSBR such that the fj-th sub-band includes the starting frequency for SBR, it is advantageous to set the type of prediction coefficients as follows:
[0062] In other words, in one embodiment, prediction coefficients for the lower sub-bands are complex values, while prediction coefficients for higher sub-bands are real values.
[0063] Second, in one embodiment, the strategy of the computation of the matrices A(k, fj) is adapted to their types. In particular, for low frequency sub-bands fj, 1 ≤ j ≤ jSBR, which are not affected by the SBR, it is possible to determine the non-zero elements of A(k, fj) by minimizing the Euclidean norm of the error between x̃(k - 1; k; fj) and its predicted version x̃P(k - 1; k; fj). The perceptual coder 31 defines and provides jSBR (not shown). In this way, phase relationships of the involved signals are explicitly exploited for prediction. For sub-band groups, the Euclidean norm of the prediction error over all directional signals of the group should be minimized (i.e. least square prediction error). For high frequency sub-bands fj, jSBR ≤ j ≤ F, which are affected by SBR, the above mentioned criterion is not reasonable, since the phases of the reconstructed sub-band coefficient sequences of the truncated HOA component
cannot be assumed to even rudimentary resemble that of the original sub-band coefficient sequences.
[0064] In this case, one solution is to disregard the phases and, instead, concentrate only on the signal powers for prediction. A reasonable criterion for the determination of the prediction coefficients is to minimize the following error
where the operation |·|2 is assumed to be applied to the matrices element-wise. In other words, the prediction coefficients are chosen such that the sum of the powers of all weighted sub-band or sub-band group coefficient sequences of the truncated HOA component best approximates the power of the directional sub-band signals. In this case, Nonnegative Matrix Factorization (NMF) techniques (see e.g. [8]) can be used to solve this optimization problem and obtain the prediction coefficients of the prediction matrices A(k, fj), j = 1, ..., F. These matrices are then provided to the Perceptual and Source Encoding stage 30.
Perceptual and source encoding
[0065] After the above-described spatial HOA coding, the resulting gain adapted transport signals for the (k - 1)-th frame, zi(k - 1), i = 1, ..., I, are coded to obtain their coded representations
(k - 1). This is performed by a Perceptual Coder 31 at the Perceptual and Source Encoding stage 30 shown in Fig.3. Further, the information contained in the sets
(k, fj), j = 1, ..., F, the prediction coefficients matrices A(k, fj) ∈
the gain control parameters ei(k - 1) and βi(k - 1), i = 1, ..., I, and the assignment vector vA(k - 1) are subjected to source encoding to remove redundancy for an efficient storage or transmission. This is performed in a Side Information Source Coder 32. The resulting coded representation
(k - 1) is multiplexed in a multiplexer 33 together with the coded transport signal representations
(k - 1), i = 1, ..., I, to provide the final coded frame
(k - 1).
[0066] Since, in principle, the source coding of the gain control parameters and the assignment can be carried out similar to [9], the present description concentrates on the coding of the directions and prediction parameters only, which is described in detail in the following.
Coding of directions
[0067] For the coding of the individual sub-band directions, the irrelevancy reduction according to the above description can be exploited to constrain the individual sub-band directions to be chosen. As already mentioned, these individual sub-band directions are chosen not out of all possible test directions ΩTEST,q, q = 1, ..., Q, but rather out of a small number of candidates determined on each frame of the full-band HOA representation. Exemplarily, a possible way for the source coding of the sub-band directions is summarized in the following Algorithm 1.
| Algorithm 1 Coding of sub-band directions |
[0068] In a first step of the Algorithm 1, the set
(k) of all full-band direction candidates that do actually occur as sub-band directions is determined, i.e.
[0069] The number of elements of this set, denoted by NoOfGlobalDirs(k), is the first part of the coded representation of the directions. Since
(k) is a subset of
(k) by definition, NoOfGlobalDirs(k) can be coded with ┌log2(D)┐ bits. To clarify the further description, the directions in the set
(k) are denoted by ΩFB,d(k), d = 1, ..., NoOfGlobalDirs(k), i.e.
[0070] In a second step, the directions in the set
(k) are coded by means of the indices q = 1, ..., Q of possible test directions ΩTEST,q, here referred to as grid. For each direction ΩFB,d(k), d = 1, ..., NoOfGlobalDirs(k), the respective grid index is coded in the array element GlobalDirGridIndices(k)[d] having a size of ┌log2(Q)┐ bits. The total array GlobalDirGridIndices(k) representing all coded full-band directions consists of NoOfGlobalDirs(k) elements.
[0071] In a third step, for each sub-band or sub-band group fj, j = 1, ..., F, the information whether the d-th directional sub-band signal (d = 1,..., DSB) is active or not, i.e. if d ∈
(k, fj), is coded in the array element bSubBandDirIsActive(k, fj)[d]. The total array bSubBandDirIsActive(k, fj) consists of DSB elements. If d ∈
(k, fj), the respective sub-band direction ΩSB,d(k, fj) is coded by means of the index i of the respective full-band direction ΩFB,i(k) into the array RelDirIndices(k, fj) consisting of DSB(k, fj) elements.
[0072] To show the efficiency of this direction encoding method, a maximum data rate for the coded representation of the directions according to the above example is calculated:
F = 10 sub-bands, DSB(k, fj) = DSB = 4 directions per sub-band, Q = 900 potential test directions and a frame rate of 25 frames per second are assumed. With the conventional coding method, the required data rate was 10 kbit/s. With the improved coding method according to one embodiment, if the number of full-band directions is assumed to be NoOfGlobalDirs(k) = D = 8, then D · ┌log2 (Q)┐ = 80 bits are needed per frame to code GlobalDirGridIndices(k), DSB · F = 40 bits to code bSubBandDirIsActive(k, fj), and
DSB · F · ┌log2(NoOfGlobalDirs(k))┐ = 120 bits to code RelDirIndices(k, fj). This results in a data rate of 240 bits/frame · 25frames/s = 6 kbit/s, which is distinctly smaller than 10 kbit/s. Even for a greater number NoOfGlobalDirs(k) = D = 16 of full-band directions, a data rate of only 7 kbit/s is sufficient.
Coding of prediction coefficient matrices
[0073] For the coding of the prediction coefficient matrices, the fact can be exploited that there is a high correlation between the prediction coefficients of successive frames due to the smoothness of the direction trajectories and consequently the directional sub-band signals. Further, there is a relatively high number of (DSB(k, fj) · MC,ACT(k - 1)) potential non-zero-elements per frame for each prediction coefficient matrix A(k, fj), where MC,ACT(k - 1) denotes the number of elements in the set
(k - 1). In total, there are F matrices to be coded per frame if no sub-band groups are used. If sub-band groups are used, there are correspondingly less than F matrices to be coded per frame.
[0074] In one embodiment, in order to keep the number of bits for each prediction coefficient low, each complex valued prediction coefficient is represented by its magnitude and its angle, and then the angle and the magnitude are coded differentially between successive frames and independently for each particular element of the matrix A(k, fj). If the magnitude is assumed to be within the interval [0,1], the magnitude difference lies within the interval [-1,1]. The difference of angles of complex numbers may be assumed to lie within the interval [-π,π]. For the quantization of both, magnitude and angle difference, the respective intervals can be subdivided into e.g. 2NQ sub-intervals of equal size. A straight forward coding then requires NQ bits for each magnitude and angle difference. Further, it has been found out experimentally that due to the above mentioned correlation between the prediction coefficients of successive frames, the occurrence probabilities of the individual differences are highly non-uniformly distributed. In particular, small differences in the magnitudes as well as in the angles occur significantly more frequently than bigger ones. Hence, a coding method that is based on the a priori probabilities of the individual values to be coded, like e.g. Huffman coding, can be exploited to reduce the average number of bits per prediction coefficient significantly. In other words, it has been found that it is usually advantageous to differentially encode magnitude and phase of the values in the prediction matrix A(k, fj), instead of their real and imaginary portions. However, there may appear circumstances under which the usage of real and imaginary portions is acceptable.
[0075] In one embodiment, special access frames are sent in certain intervals (application specific, e.g. once per second) that include the non-differentially coded matrix coefficients. This allows a decoder to re-start a differential decoding from these special access frames, and thus enables a random entry for the decoding.
[0076] In the following, decompression of a low bit rate compressed HOA representation as constructed above is described. Also the decompression works frame-wise.
[0077] In principle, a low bit rate HOA decoder, according to an embodiment, comprises counterparts of the above-described low bit rate HOA encoder components, which are arranged in reverse order. In particular, the low bit rate HOA decoder can be subdivided into a perceptual and source decoding part as depicted in Fig.4, and a spatial HOA decoding part as illustrated in Fig.6.
Perceptual and source decoding
[0078] Fig.4 shows a Perceptual and Side Info Source Decoder 40, in one embodiment. In the Perceptual and Side Info Source Decoder 40, the low bit rate compressed HOA bit stream B is first de-multiplexed 41, which results in a perceptually coded representation of the I signals
i = 1, ..., I, and the coded side information
describing how to create a HOA representation thereof. Successively, a perceptual decoding of the I signals and a decoding of the side information is performed.
[0079] A Perceptual Decoder 42 decodes the I signals
(k), i = 1, ..., I into the perceptually decoded signals ẑi(k), i = 1,...,I.
[0080] A Side Information Source decoder 43 decodes the coded side information
into the tuple sets
(k + 1, fj), j = 1, ..., F, the prediction coefficient matrices A(k + 1, fj) for each sub-band or sub-band group fj (j = 1, ..., F), gain correction exponents ei(k) and gain correction exception flags βi(k), and assignment vector vAMB,ASSIGN(k).
[0081] Algorithm 2 summarizes exemplarily how to create the tuple sets
(k, fj), j = 1,..., F, from the coded side information
The decoding of the sub-band directions is described in detail in the following.
| Algorithm 2 Decoding of sub-band directions |
[0082] First, the number of full-band directions NoOfGlobalDirs(k) is extracted from the coded side information
As described above, these are also used as sub-band directions. It is coded with ┌log2(D)┐ bits.
[0083] In a second step, the array GlobalDirGridIndices(k) consisting of NoOfGlobalDirs(k) elements is extracted, each element being coded by ┌log2(Q)┐ bits. This array contains the grid indices that represent the full-band directions ΩFB,d(k),
d = 1, ..., NoOfGlobalDirs(k), such that
[0084] Then, for each sub-band or sub-band group fj, j = 1, ..., F, the array bSubBandDirIsActive(k, fj) consisting of DSB elements is extracted, where the d-th element bSubBandDirIsActive(k, fj)[d] indicates whether or not the d-th sub-band direction is active. Further, the total number of active sub-band directions DSB(k, fj) is computed.
[0085] Finally, the set
(k, fj) of tuples is computed for each sub-band or sub-band group fj, j = 1, ..., F. It consists of the indices d ∈
(k, fj) ⊆ {1,,DSB} that identify the individual (active) sub-band direction trajectories, and the respective estimated directions ΩSB,d(k, fj).
[0086] Next, the prediction coefficient matrices A(k + 1, fj) for each sub-band or sub-band group fj, j = 1, ..., F are reconstructed from the coded frame
(k). In one embodiment, the reconstruction comprises the following steps per sub-band or sub-band group fj: First, the angle and magnitude differences of each matrix coefficient are obtained by entropy decoding. Then, the entropy decoded angle and magnitude differences are rescaled to their actual value ranges, according to the number of bits NQ used for their coding. Finally, the current prediction coefficient matrix A(k + 1, fj) is built by adding the reconstructed angle and magnitude differences to the coefficients of the latest coefficient matrix A(k, fj), i.e. the coefficient matrix of the previous frame.
[0087] Thus, the previous matrix A(k, fj) has to be known for the decoding of a current matrix A(k + 1, fj). In one embodiment, in order to enable a random access, special access frames are received in certain intervals that include the non-differentially coded matrix coefficients to re-start the differential decoding from these frames.
[0088] The Perceptual and Side Info Source Decoder 40 outputs the perceptually decoded signals ẑi(k), i = 1, ..., I, tuple sets
(k + 1, fj), j = 1, ..., F, prediction coefficient matrices A(k + 1, fj), gain correction exponents ei(k), gain correction exception flags βi(k) and assignment vector vAMB,ASSIGN(k) to a subsequent Spatial HOA decoder 50.
Spatial HOA decoding
[0089] Fig.5 shows an exemplary Spatial HOA decoder 50, in one embodiment. The spatial HOA decoder 50 creates from the I signals ẑi(k), i = 1,...,I, and the above-described side information provided by the Side Information Decoder 43 a reconstructed HOA representation. The individual processing units within the spatial HOA decoder 50 are described in detail in the following.
Inverse Gain Control
[0090] In the Spatial HOA decoder 50, the perceptually decoded signals ẑi(k), i = 1,...,I, together with the associated gain correction exponent ei(k) and gain correction exception flag βi(k), are first input to one or more Inverse Gain Control processing blocks 51. The Inverse Gain Control processing blocks provide gain corrected signal frames ŷi(k), i = 1,...,I. In one embodiment, each of the I signals ẑi(k) is fed into a separate Inverse Gain Control processing block 51, as in Fig.5, so that the i-th Inverse Gain Control processing block provides a gain corrected signal frame ŷi(k). A more detailed description of the Inverse Gain Control is known from e.g. [9], Section 11.4.2.1.
[0092] In a Truncated HOA Reconstruction block 52, the I gain corrected signal frames ŷi(k), i = 1,...,I, are redistributed (i.e. reassigned) to a HOA coefficient sequence matrix, according to the information provided by the assignment vector vAMB,ASSIGN(k), so that the truncated HOA representation ĈT(k) is reconstructed. The assignment vector vAMB,ASSIGN(k) comprises I components that indicate for each transmission channel which coefficient sequence of the original HOA component it contains. Further, the elements of the assignment vector form a set
(k) of the indices, referring to the original HOA component, of all the received coefficient sequences for the k-th frame
[0094] First, the individual components ĉI,n(k), n = 1, ..., O, of the decoded intermediate representation
are either set to zero or replaced by a corresponding component of the gain corrected signal frames ŷi(k), depending on the information in the assignment vector, i.e.
[0095] This means, as mentioned above, that the i-th element of the assignment vector, which is n in eq.(26), indicates that the i-th coefficient ŷi(k) replaces ĉI,n(k) in the n-th line of the decoded intermediate representation matrix ĈI(k).
[0096] Second, a re-correlation of the first OMIN signals within ĈI(k) is carried out by applying to them the inverse spatial transform, providing the frame
where the mode matrix ΨMIN is as defined in eq.(6). The mode matrix depends on given directions that are predefined for each OMIN or NMIN respectively, and can thus be constructed independently both at the encoder and decoder. Also OMIN (or NMIN) is predefined by convention.
[0097] Finally, the reconstructed truncated HOA representation ĈT(k) is composed from the re-correlated signals ĈT,MIN(k) and the signals of the intermediate representation ĉI,n(k), n = OMIN + 1, ..., O, according to
Analysis Filter Banks
[0098] To further compute the second HOA component, which is represented by predicted directional sub-band signals, each frame ĉT,n(k), n = 1, ..., O, of an individual coefficient sequence n of the decompressed truncated HOA representation ĈT(k) is first decomposed in one or more Analysis Filter Banks 53 into frames of individual sub-band signals
For each sub-band fj, j = 1, ..., F, the frames of the sub-band signals of the individual HOA coefficient sequences may be collected into the sub-band HOA representation
as
[0099] The one or more Analysis Filter Banks 53 applied at the HOA spatial decoding stage are the same as those one or more Analysis Filter Banks 15 at the HOA spatial encoding stage, and for sub-band groups the grouping from the HOA spatial encoding stage is applied. Thus, in one embodiment, grouping information is included in the encoded signal. More details about grouping information is provided below.
[0100] In one embodiment, a maximum order NMAX is considered for the computation of the truncated HOA representation at the HOA compression stage (see above, near eq.(4)), and the application of the HOA compressor's and decompressor's Analysis Filter Banks 15, 53 is restricted to only those HOA coefficient sequences ĉT,n(k) with indices n = 1, ..., OMAX. The sub-band signal frames
with indices n = OMAX + 1, ..., O can then be set to zero.
Synthesis of directional sub-band HOA representation
[0101] For each sub-band or sub-band group, directional sub-band or sub-band group HOA representations
are synthesized in one or more Directional Sub-band Synthesis blocks 54. In one embodiment, in order to avoid artifacts due to changes of the directions and prediction coefficients between successive frames, the computation of the directional sub-band HOA representation is based on the concept of overlap add.
[0102] Hence, in one embodiment, the HOA representation
of active directional sub-band signals related to the fj-th sub-band, j = 1,..., F, is computed as the sum of a faded out component and a faded in component:
[0103] In a first step, to compute the two individual components, the instantaneous frame of all directional sub-band signals x̃I(k1;k; fj) related to the prediction coefficients matrices A(k1, fj) for frames k1 ∈ {k, k + 1} and the truncated sub-band HOA representation
for the k-th frame is computed by
[0104] For sub-band groups, the HOA representations of each group
are multiplied by a fixed matrix A(k1,fj) to create the sub-band signals x̃I(k1;k;fj) of the group.
[0105] In a second step, the instantaneous sub-band HOA representation
d ∈
(k,fj), j = 1, ..., F, of the directional sub-band signal x̃I,d(k1;k;fj) with respect to the direction ΩSB,d(k,fj) is obtained as
where
denotes the mode vector (as the mode vectors in eq.(7)) with respect to the direction ΩSB,d(k,fj). For sub-band groups, eq. (32) is performed for all signals of the group, where the matrix ψ(ΩSB,d(k,fj)) is fixed for each group.
[0106] Assuming the matrices
to be composed of their samples by
the sample values of the faded out and faded in components of the HOA representation of active directional sub-band signals are finally determined by
where the vector
represents an overlap add window function. An example for the window function is given by the periodic Hann window, the elements of which being defined by
Sub-band HOA Composition
[0107] For each sub-band or sub-band group fj, j = 1, ..., F, the coefficient sequences
n = 1, ..., 0, of the decoded sub-band HOA representation
are either set to that of the truncated HOA representation
if it was previously transmitted, or else to that of the directional HOA component
provided by one of the Directional Sub-band Synthesis blocks 54, i.e.
[0108] This sub-band composition is performed by one or more Sub-band Composition blocks 55. In an embodiment, a separate Sub-band Composition block 55 is used for each sub-band or sub-band group, and thus for each of the one or more Directional Sub-band Synthesis blocks 54. In one embodiment, a Directional Sub-band Synthesis block 54 and its corresponding Sub-band Composition block 55 are integrated into a single block.
Synthesis Filter Banks
[0109] In a final step, the decoded HOA representation is synthesized from all the decoded sub-band HOA representations
The individual time domain coefficient sequences
of the decompressed HOA representation Ĉ(k), are synthesized from the corresponding sub-band coefficient sequences
by one or more Synthesis Filter Banks 56, which finally outputs the decompressed HOA representation Ĉ(k).
[0110] Note that the synthesized time domain coefficient sequences usually have a delay due to successive application of the analysis and synthesis filter banks 53, 56.
[0111] Fig.8 shows exemplarily, for a single frequency subband f1 , a set of active direction candidates, their chosen trajectories and corresponding tuple sets. In a frame k, four directions are active in a frequency subband f1. The directions belong to respective trajectories T1, T2, T3 and T5. In previous frames k-2 and k-1, different directions were active, namely T1, T2, T6 and T1-T4, respectively. The set of active directions MDIR(K) in the frame k relates to the full band and comprises several active direction candidates, e.g. MDIR(k)={Ω3, Ω8, Ω52, Ω101,Ω229, Ω446, Ω581}. Each direction can be expressed in any way, e.g. by two angles or as an index of a predefined table. From the set of active full-band directions, those directions that are actually active in a subband and their corresponding trajectories are collected, separately for each frequency subband, in the tuple sets MDIR(k,fj), j=1,...,F. For example, in the first frequency subband of frame k, active directions are Ω3, Ω52, Ω229 and Ω581, and their associated trajectories are T3, T1, T2 and T5 respectively. In the second frequency subband f2, active directions are exemplarily only Ω52 and Ω229, and their associated trajectories are T1 and T2 respectively.
[0112] The following is a portion of a coefficient matrix of an exemplary truncated HOA representation CT(k), corresponding to the coefficient sequences in an exemplary set IC,ACT(k) = {1,2,4,6}:
[0113] According to IC,ACT(k), only coefficients of the rows 1, 2, 4 and 6 are not set to zero (nevertheless, they may be zero, depending on the signal). Each column of the matrix CT(k) refers to a sample, and each row of the matrix is a coefficient sequence. The compression comprises that not all coefficient sequences are encoded and transmitted, but only some selected coefficient sequences, namely those whose indices are included in IC,ACT(k) and the assignment vector vA(k) respectively. At the decoder, the coefficients are decompressed and positioned into the correct matrix rows of the reconstructed truncated HOA representation. The information about the rows is obtained from the assignment vector vAMB,ASSIGN(k), which provides additionally also the transport channels that are used for each transmitted coefficient sequence. The remaining coefficient sequences are filled with zeros, and later predicted from the received (usually non-zero) coefficients according to the received side information, e.g. the subband or subband group related prediction matrices and directions.
Sub-band grouping
[0114] In one embodiment, the used subbands have different bandwidths adapted to the psycho-acoustic properties of human hearing. Alternatively, a number of subbands from the Analysis Filter Bank 53 are combined so as to form an adapted filter bank with subbands having different bandwidths. A group of adjacent subbands from the Analysis Filter Bank 53 is processed using the same parameters. If groups of combined subbands are used, the corresponding subband configuration applied at the encoder side must be known to the decoder side. In an embodiment, configuration information is transmitted and is used by the decoder to set up its synthesis filter bank. In an embodiment, the configuration information comprises an identifier for one out of a plurality of predefined known configurations (e.g. in a list).
[0115] In another embodiment, the following flexible solution that reduces the required number of bits for defining a subband configuration is used. For an efficient encoding of subband configuration, data of the first, penultimate and last subband groups are treated differently than the other subband groups. Further, subband group bandwidth difference values are used in the encoding. In principle, the subband grouping information coding method is suited for coding subband configuration data for subband groups valid for one or more frames of an audio signal, wherein each subband group is a combination of one or more adjacent original subbands and the number of original subbands is predefined. In one embodiment, the bandwidth of a following subband group is greater than or equal to the bandwidth of a current subband group. The method includes coding a number of NSB subband groups with a fixed number of bits representing NSB - 1, and if NSB > 1, coding for a first subband group g1 a bandwidth value BSB[1] with a unary code representing BSB[1] - 1.If NSB = 3, a bandwidth difference value ΔBSB[2] = BSB[2] - BSB[1] with a fixed number of bits is coded for a second subband group g2. If NSB > 3, a corresponding number of bandwidth difference values ΔBSB[g] = BSB[g] - BSB[g - 1] is coded for the subband groups g2,...,gNSB-2 with a unary code, and a bandwidth difference value ΔBsB[NSB - 1] = BSB[N]SOB - 11 - BSB[Nsb - 2] with a fixed number of bits is coded for the last subband group gNSB-1. A bandwidth value for a subband group is expressed as a number of adjacent original subbands. For the last subband group gSB, no corresponding value needs to be included in the coded subband configuration data.
[0116] Fig.9 shows a generalized block diagram of the HOA encoding path of a conventional MPEG-H 3D audio encoder. Two types of predominant sound signals are extracted: directional signals in a Directional Sound Extraction block DSE and vector-based signals VVec in a VVec Sound Extraction block VSE. The vector belonging to a vector-based signal VVec (V-vector) represents the spatial distribution of the soundfield for the corresponding vector-based signal. Further, also an ambiance component is encoded in a Calculator for Residuum/Ambience CRA, whereby any one or both or none of the output data from the Directional Sound Extraction block DSE and the VVec Sound Extraction block VSE can be used. The ambience signal is subjected to Spatial Resolution Reduction block SRR, partial decorrelation PD and gain control GCA. The blocks within the box are controlled by the Sound Scene Analysis SSA. Before being fed into the Universal Speech &Audio encoder USAC3D, also the predominant sound signals are processed by respective gain control blocks GCD,GCV. Finally the USAC3D encoder ENCc&HEPC packs the HOA spatial side information into the HOA extension payload.
[0117] Fig.10 shows an improved audio encoder as usable in MPEG, according to one embodiment. The disclosed technology amends the current MPEG-H 3D Audio system in a way that the bit stream for low bandwidth is a real superset of the known MPEG-H 3D Audio format. Compared to Fig.9, in the Sound Scene Analysis SSA a path is added that comprises two new blocks. These are a QMF Analysis Filter bank QAC, which is applied to ambiance signals, and a Directional Subband Calculation block DSCC for calculation of parameters of directional subband signals. These parameters allow for synthesizing directional signals based on the transmitted ambiance signals. Additionally, parameters are calculated which allow for reproducing missing ambiance signals. The side information parameters for the synthesis process are handed over to the USAC3D encoder ENC&HEP, which packs them into the HOA extension payload of the compressed output signal HOAC,O. Advantageously, the compression is more efficient than conventional compression as achieved with the arrangement of Fig.9.
[0118] Fig.11 shows a generalized block diagram of a conventional MPEG-H 3D Audio decoder. First, the HOA side information is extracted from the compressed input bitstream HOAC,l and a USAC3D and HOA Extension Payload decoder DECC&HEPC reproduces the transmission channels waveform signals. These are fed into the corresponding inverse gain control blocks IGCD, IGCV, IGCA. Here, the normalization applied in the encoder is reversed. The corresponding transmission channels are used together with the side information to synthesize the predominant sound signals (directional and/or vector-based) in a HOA Directional Sound Synthesis block DSS and/or a VVec Sound Synthesis block VSS respectively. In the third path, the ambiance component is reproduced by Inverse Partial Decorrelation IPD and HOA Ambience Synthesis HAS blocks. The following HOA Composition block HCC combines the predominant sound components and the ambiance to build the decoded HOA signal. This is fed into the HOA renderer HR to produce the output signal HOA'D,O, ie. the final loudspeaker feeds.
[0119] Fig.12 shows an improved audio decoder as usable in MPEG, according to one embodiment. As in the encoder, a path is added. It comprises a decoder side QMF Analysis block QAD for calculation of subband signals and a Directional Subband signal Synthesis block DSCD for the synthesis of the parametrically encoded directional subband signals. The calculated subband signals are used together with the corresponding transmitted side information to synthesize a HOA representation of directional signals. Afterwards, the synthesized signal component is transferred into the time domain using the QMF synthesis filter bank QS. Its output signal is additionally fed into the enhanced HOA composition block HC. The following HOA rendering block HR for providing a decoded HOA output signal HOAD,O is left unchanged.
[0120] In the following, some basic features of Higher Order Ambisonics are explained. Higher Order Ambisonics (HOA) is based on the description of a sound field within a compact area of interest, which is assumed to be free of sound sources. In that case the spatiotemporal behavior of the sound pressure p(t,x) at time t and position x within the area of interest is physically fully determined by the homogeneous wave equation. In the following we assume a spherical coordinate system as shown in Fig.6. In this coordinate system, the x axis points to the frontal position, the y axis points to the left, and the z axis points to the top. A position in space x = (r, θ, φ)T is represented by a radius r > 0 (i.e. the distance to the coordinate origin), an inclination angle θ ∈ [0,π] measured from the polar axis z (!) and an azimuth angle φ ∈ [0,2π[ measured counter-clockwise in the x - y plane from the x axis. Further, (·)T denotes the transposition.
[0121] Then, it can be shown [11] that the Fourier transform of the sound pressure with respect to time denoted by
(·), i.e.,
with ω denoting the angular frequency and i indicating the imaginary unit, may be expanded into the series of Spherical Harmonics according to
[0122] In eq.(42), cs denotes the speed of sound and k denotes the angular wave number, which is related to the angular frequency ω by
Further, jn(·) denote the spherical Bessel functions of the first kind and
denote the real valued Spherical Harmonics of order n and degree m, which are defined above. The expansion coefficients
only depend on the angular wave number k. Note that it has been implicitly assumed that sound pressure is spatially band-limited. Thus, the series is truncated with respect to the order index n at an upper limit N, which is called the order of the HOA representation.
[0123] If the sound field is represented by a superposition of an infinite number of harmonic plane waves of different angular frequencies ω and arriving from all possible directions specified by the angle tuple (θ,φ), it can be shown [10] that the respective plane wave complex amplitude function C(ω,θ,φ) can be expressed by the following Spherical Harmonics expansion
where the expansion coefficients
are related to the expansion coefficients
by
[0124] Assuming the individual coefficients
to be functions of the angular frequency ω, the application of the inverse Fourier transform (denoted by
(·)) provides time domain functions
for each order n and degree m. These time domain functions are referred to as continuous-time HOA coefficient sequences here, which can be collected in a single vector c(t) by
[0125] The position index of a HOA coefficient sequence
within the vector c(t) is given by n(n + 1) + 1 + m.
[0127] The final Ambisonics format provides the sampled version of c(t) using a sampling frequency fS as
where TS = 1/fS denotes the sampling period. The elements of c(lTS) are here referred to as discrete-time HOA coefficient sequences, which can be shown to always be real valued. This property obviously also holds for the continuous-time versions
Definition of real valued Spherical Harmonics
[0128] The real valued spherical harmonics
(assuming SN3D normalization [1, Ch.3.1]) are given by
with
[0129] The associated Legendre functions Pn,m(x) are defined as
with the Legendre polynomial Pn(x) and, unlike in [11], without the Condon-Shortley phase term (-1)m.
[0130] In one embodiment, a method for frame-wise determining and efficient encoding of directions of dominant directional signals within subbands or subband groups of a HOA signal representation (as obtained from a complex valued filter bank) comprises for each current frame k: determining a set MDIR(K) of full band direction candidates in the HOA signal, a number of elements NoOfGlobalDirs in the set MDIR(K) and a number D(k)=log2(NoOfGlobalDirs) required for encoding the number of elements, wherein each full band direction candidate has a global index q (q ∈ [1, ..., Q]) relating to a predefined full set of Q possible directions,
[0131] for each subband or subband group j of the current frame k, determining which directions of the full band direction candidates in the set MDIR(K) occur as active subband directions, determining a set MFB(k) of used full band direction candidates (all contained in the set MDIR(K) of full band direction candidates in the HOA signal) that occur as active subband directions in any of the subbands or subband groups, and a number NoOfGlobalDirs(k) of elements in the set MFB(k) of used full band direction candidates, and for each subband or subband group j of the current frame k: determining which directions of up to d (d ∈ [1, ..., D]) directions among the full band direction candidates in the set MDIR(K) are active subband directions, determining for each of the active subband directions a trajectory and a trajectory index, and assigning the trajectory index to each active subband direction, and
encoding each of the active subband directions in the current subband or subband group j by a relative index with D(k) bits.
[0132] In one embodiment, a computer readable medium has stored thereon executable instructions to cause a computer to perform this method for frame-wise determining and efficient encoding of directions of dominant directional signals.
[0133] Further, in one embodiment, a method for decoding of directions of dominant directional signals within subbands of a HOA signal representation comprises steps of receiving indices of a maximum number of directions D for a HOA signal representation to be decoded, reconstructing directions of a maximum number of directions D of the HOA signal representation to be decoded, receiving indices of active direction signals per subband, reconstructing active directions per subband from the reconstructed directions D of the HOA signal representation to be decoded and the indices of active direction signals per subband, predicting directional signals of subbands, wherein the predicting of a directional signal in a current frame of a subband comprises determining directional signals of a preceding frame of the subband, and wherein a new directional signal is created if the index of the directional signal was zero in the preceding frame and is non-zero in the current frame, a previous directional signal is cancelled if the index of the directional signal was non-zero in the preceding frame and is zero in the current frame, and a direction of a directional signal is moved from a first to a second direction if the index of the directional signal changes from the first to the second direction.
[0134] In one embodiment, as shown in Fig.1 and Fig.3 and discussed above, an apparatus for encoding frames of an input HOA signal having a given number of coefficient sequences, where each coefficient sequence has an index, comprises at least one hardware processor and a non-transitory, tangible, computer-readable storage medium tangibly embodying at least one software component that when executing on the at least one hardware processor causes
computing 11 a truncated HOA representation CT(k) having a reduced number of non-zero coefficient sequences,
determining 11 a set of indices of active coefficient sequences IC,ACT(k) that are included in the truncated HOA representation,
estimating 16 from the input HOA signal a first set of candidate directions MDIR(k); dividing 15 the input HOA signal into a plurality of frequency subbands f1, ..., fF, wherein coefficient sequences
of the frequency subbands are obtained,
estimating 16 for each of the frequency subbands a second set of directions MDIR(k,f1), ..., MDIR(k,fF), wherein each element of the second set of directions is a tuple of indices with a first and a second index, the second index being an index of an active direction for a current frequency subband and the first index being a trajectory index of the active direction, wherein each active direction is also included in the first set of candidate directions MDIR(K) of the input HOA signal,
for each of the frequency subbands, computing 17 directional subband signals
from the coefficient sequences
of the frequency subband according to the second set of directions MDIR(k,f1),...,MDIR(k,fF) of the respective frequency subband,
for each of the frequency subbands, calculating 18 a prediction matrix A(k,f1),..., A(k,fF) adapted for predicting the directional subband signals
from the coefficient sequences
of the frequency subband using the set of indices of active coefficient sequences IC,ACT(k) of the respective frequency subband, and
encoding the first set of candidate directions MDIR(k), the second set of directions MDIR(k,f1),..., MDIR(k,fF), the prediction matrices A(k,f1),...,A(k,fF) and the truncated HOA representation CT(k).
[0135] In one embodiment, as shown in Fig.4 and Fig.5 and discussed above, an apparatus for decoding a compressed HOA representation comprises at least one hardware processor and a non-transitory, tangible, computer-readable storage medium tangibly embodying at least one software component that when executing on the at least one hardware processor causes extracting 41,42,43 from the compressed HOA representation a plurality of truncated HOA coefficient sequences ẑ1(k), ... , ẑI(k), an assignment vector vAMB,ASSIGN(k) indicating or containing sequence indices of said truncated HOA coefficient sequences, subband related direction information MDIR(k+1,f1),..., MDIR(k+1,fF), a plurality of prediction matrices A(k+1,f1),...,A(k+1,fF), and gain control side information e1(k),β1(k),...,eI(k),βI(k);
reconstructing 51,52 a truncated HOA representation ĈT(k) from the plurality of truncated HOA coefficient sequences ẑ1(k),...,ẑI(k), the gain control side information e1(k),β1(k), ...,eI(k),βI(k) and the assignment vector vAMB,ASSIGN(k),
decomposing in Analysis Filter banks 53 the reconstructed truncated HOA representation ĈT(k) into frequency subband representations
for a plurality of F frequency subbands,
synthesizing 54 in Directional Subband Synthesis blocks 54 for each of the frequency subband representations a predicted directional HOA representation
...,
from the respective frequency subband representation
of the reconstructed truncated HOA representation, the subband related direction information MDIR(k+1,f1),...,MDIR(k+1,fF) and the prediction matrices A(k+1,f1),...,A(k+1,fF), composing 55 in Subband Composition blocks 55 for each of the Ffrequency subbands a decoded subband HOA representation
with coefficient sequences
that are either obtained from coefficient sequences of the truncated HOA representation
if the coefficient sequence has an index n that is included in the assignment vector vAMB,ASSIGN(k), or otherwise obtained from coefficient sequences of the predicted directional HOA component
provided by one of the Directional Subband Synthesis blocks 54, and synthesizing in Synthesis Filter banks 56 the decoded subband HOA representations
to obtain the decoded HOA representation Ĉ(k).
[0136] In one embodiment, an apparatus 10 for encoding frames of an input HOA signal having a given number of coefficient sequences, where each coefficient sequence has an index, comprises a computation and determining module 11 configured to compute a truncated HOA representation CT(k) having a reduced number of non-zero coefficient sequences, and further configured to determine a set of indices of active coefficient sequences IC,ACT(k) included in the truncated HOA representation;
an Analysis Filter bank module 15 configured to divide the input HOA signal into a plurality of frequency subbands f1,...,fF, wherein coefficient sequences
...,
of the frequency subbands are obtained;
a Direction Estimation module 16 configured to estimate from the input HOA signal a first set of candidate directions MDIR(k), and further configured to estimate for each of the frequency subbands a second set of directions MDIR(k,f1), ..., MDIR(k,fF), wherein each element of the second set of directions is a tuple of indices with a first and a second index, the second index being an index of an active direction for a current frequency subband and the first index being a trajectory index of the active direction, wherein each active direction is also included in the first set of candidate directions MDIR(K) of the input HOA signal; at least one Directional Subband Computation module 17 configured to compute, for each of the frequency subbands, directional subband signals X̃I(k - 1, k, f1), ..., X̃(k - 1, k,fF) from the coefficient sequences
of the frequency subband according to the second set of directions MDIR(k,f1),...,MDIR(k,fF) of the respective frequency subband; at least one Directional Subband Prediction module 18 configured to calculate, for each of the frequency subbands, a prediction matrix A(k,f1),..., A(k,fF) adapted for predicting the directional subband signals X̃(k - 1, k,f1),..., X̃I(k - 1, k,fF) from the coefficient sequences
of the frequency subband using the set of indices of active coefficient sequences IC,ACT(k) of the respective frequency subband; and an encoding module 30 configured to encode the first set of candidate directions MDIR(k), the second set of directions MDIR(k,f1),..., MDIR(k,fF), the prediction matrices A(k,f1),..., A(k,fF) and the truncated HOA representation CT(k).
[0137] In one embodiment, the apparatus further comprises a Partial Decorrelator 12 configured to partially decorrelate the truncated HOA channel sequences; a Channel Assignment module 13 configured to assigning the truncated HOA channel sequences y1(k),..., yI(k) to transport channels; and at least one Gain Control unit 14 configured to perform gain control on the transport channels, wherein gain control side information ei(k - 1),
βi(k - 1) for each transport channel is generated.
[0138] In one embodiment, the encoding module 30 comprises a Perceptual Encoder 31 configured to encode the gain controlled truncated HOA channel sequences z1(k),...,zI(k); a Side Information Source Coder 32 configured to encode the gain control side information ei(k - 1), βi(k - 1), the first set of candidate directions MDIR(k), the second set of directions MDIR(k,f1),..., MDIR(k,fF) and the prediction matrices A(k,f1),...,A(k,fF); and a Multiplexer 33 configured to multiplex the outputs of the perceptual encoder 31 and the side information source coder 32 to obtain an encoded HOA signal frame
(k - 1).
[0139] In one embodiment, an apparatus 50 for decoding a HOA signal comprises
an Extraction module 40 configured to extract from the compressed HOA representation a plurality of truncated HOA coefficient sequences ẑ1(k), ... , ẑI(k), an assignment vector vAMB,ASSIGN(k) indicating or containing sequence indices of said truncated HOA coefficient sequences, subband related direction information MDIR(k+1,f1),...,MDIR(k+1,fF), a plurality of prediction matrices A(k+1,f1),...,A(k+1,fF), and gain control side information e1(k),β1(k),...,eI(k),βI(k); a Reconstruction module 51,52 configured to reconstruct a truncated HOA representation ĈT(k) from the plurality of truncated HOA coefficient sequences ẑ1(k),...,ẑI(k), the gain control side information e1(k),β1(k),...,eI(k),βI(k) and the assignment vector vAMB,ASSIGN(k); an Analysis Filter bank module 53 configured to decompose the reconstructed truncated HOA representation ĈT(k) into frequency subband representations
for a plurality of F frequency subbands;
at least one Directional Subband Synthesis module 54 configured to synthesize for each of the frequency subband representations a predicted directional HOA representation
from the respective frequency subband representation
of the reconstructed truncated HOA representation, the subband related direction information MDIR(k+1,f1),...,MDIR(k+1,fF) and the prediction matrices A(k+1,f1),...,A(k+1,fF);
at least one Subband Composition module 55 configured to compose for each of the F frequency subbands a decoded subband HOA representation
with coefficient sequences
that are either obtained from coefficient sequences of the truncated HOA representation
if the coefficient sequence has an index n that is included in the assignment vector vAMB,ASSIGN(k), or otherwise obtained from coefficient sequences of the predicted directional HOA component
provided by one of the Directional Subband Synthesis module 54; and a Synthesis Filter bank module 56 configured to synthesize the decoded subband HOA representations
to obtain the decoded HOA representation Ĉ(k).
[0140] In one embodiment, the Extraction module 40 comprises at least a Demultiplexer 41 for obtaining an encoded side information portion and a perceptually coded portion that comprises encoded truncated HOA coefficient sequences
a Perceptual Decoder 42 configured to perceptually decode s42 the encoded truncated HOA coefficient sequences
to obtain the truncated HOA coefficient sequences ẑ1(k),...,ẑI(k); and a Side Information Source Decoder 43 configured to decode (s43) the encoded side information portion to obtain the subband related direction information MDIR(k+1,f1),..., MDIR(k+1,fF), prediction matrices A(k+1,f1),...,A(k+1,fF), gain control side information e1(k),β1(k),...,eI(k),βI(k) and assignment vector vAMBASSIGN(k).
[0141] Fig.13 shows a flow-chart of a low bit-rate encoding method, in one embodiment. The method for low bit-rate encoding of frames of an input HOA signal having a given number of coefficient sequences, where each coefficient sequence has an index, comprises computing s110 a truncated HOA representation CT(k) having a reduced number of non-zero coefficient sequences, determining s111 a set of indices of active coefficient sequences IC,ACT(k) that are included in the truncated HOA representation, estimating s16 from the input HOA signal a first set of candidate directions MDIR(k), dividing s15 the input HOA signal into a plurality of frequency subbands f1,...,fF, wherein coefficient sequences C̃(k - 1,k,f1),...,C̃(k - 1,k,fF) of the frequency subbands are obtained, estimating s161 for each of the frequency subbands a second set of directions MDIR(k,f1), ..., MDIR(k,fF), wherein each element of the second set of directions is a tuple of indices with a first and a second index, the second index being an index of an active direction for a current frequency subband and the first index being a trajectory index of the active direction, wherein each active direction is also included in the first set of candidate directions MDIR(K) of the input HOA signal,
for each of the frequency subbands, computing s17 directional subband signals X̃(k - 1,k,f1),...,X̃(k - 1,k,fF) from the coefficient sequences
C̃(k - 1, k, fF) of the frequency subband according to the second set of directions MDIR(k,f1),...,MDIR(k,fF) of the respective frequency subband,
for each of the frequency subbands, calculating s18 a prediction matrix A(k,f1),...,A(k,fF) adapted for predicting the directional subband signals X̃(k - 1,k,f1),..., X̃(k - 1, k,fF) from the coefficient sequences
of the frequency subband using the set of indices of active coefficient sequences IC,ACT(k) of the respective frequency subband, and encoding s19 the first set of candidate directions MDIR(k), the second set of directions MDIR(k,f1),..., MDIR(k,fF), the prediction matrices A(k,f1),...,A(k,fF) and the truncated HOA representation CT(k).
[0142] In one embodiment, said encoding the truncated HOA representation CT(k) comprises partial decorrelation s12 of the truncated HOA channel sequences, channel assignment s13 for assigning the truncated HOA channel sequences y1(k),..., yI(k) to transport channels, performing gain control s14 on each of the transport channels, wherein gain control side information ei(k - 1), βi(k - 1) for each transport channel is generated, encoding s31 the gain controlled truncated HOA channel sequences z1(k),...,zI(k) in a perceptual encoder 31, encoding s32 the gain control side information ei(k - 1),
βi(k - 1), the first set of candidate directions MDIR(k), the second set of directions MDIR(k,f1),...,MDIR(k,fF) and the prediction matrices A(k,f1),...,A(k,fF) in a side information source coder 32, and multiplexing s33 the outputs of the perceptual encoder 31 and the side information source coder 32 to obtain an encoded HOA signal frame
(k - 1).
[0143] In one embodiment, an apparatus for encoding frames of an input HOA signal having a given number of coefficient sequences, where each coefficient sequence has an index, comprises a processor and a memory storing instructions that, when executed, cause the apparatus to perform the steps of claim 7.
[0144] Fig.14 shows a flow-chart of a decoding method, in one embodiment. The method for decoding a low bit-rate compressed HOA representation, comprises extracting s41,s42,s43 from the compressed HOA representation a plurality of truncated HOA coefficient sequences ẑ1(k),...,ẑI(k), an assignment vector vAMB,ASSIGN(k) indicating or containing sequence indices of said truncated HOA coefficient sequences, subband related direction information MDIR(k+1,f1),..., MDIR(k+1,fF), a plurality of prediction matrices A(k+1,f1),...,A(k+1,fF), and gain control side information e1(k),β1(k),...,eI(k),βI(k), reconstructing s51,s52 a truncated HOA representation ĈT(k) from the plurality of truncated HOA coefficient sequences ẑ1(k), ... , ẑI(k), the gain control side information e1(k),β1(k),...,eI(k),βI(k) and the assignment vector vAMB,ASSIGN(k), decomposing s53 in Analysis Filter banks 53 the reconstructed truncated HOA representation (ĈT(k)) into frequency subband representations
for a plurality of Ffrequency subbands, synthesizing s54 in Directional Subband Synthesis blocks 54 for each of the frequency subband representations a predicted directional HOA representation
from the respective frequency subband representation
of the reconstructed truncated HOA representation, the subband related direction information MDIR(k+1,f1),...,MDIR(k+1,fF) and the prediction matrices A(k+1,f1),...,A(k+1,fF), composing s55 in Subband Composition blocks 55 for each of the F frequency subbands a decoded subband HOA representation
with coefficient sequences
that are either obtained from coefficient sequences of the truncated HOA representation
if the coefficient sequence has an index n that is included in the assignment vector vAMB,ASSIGN(k), or otherwise obtained from coefficient sequences of the predicted directional HOA component
provided by one of the Directional Subband Synthesis blocks 54, and synthesizing s56 in Synthesis Filter banks 56 the decoded subband HOA representations
to obtain the decoded HOA representation Ĉ(k).
[0145] In an embodiment, the extracting comprises one or more of demultiplexing s41 the compressed HOA representation to obtain a perceptually coded portion and an encoded side information portion, perceptually decoding s42 the encoded truncated HOA coefficient sequences and decoding s43 in a side information source decoder 43 the encoded side information. In an embodiment, the reconstructing a truncated HOA representation ĈT(k) from the plurality of truncated HOA coefficient sequences comprises one or more of performing inverse gain control s51 and reconstructing s52 the truncated HOA representation ĈT(k). In one embodiment, a computer readable medium has stored thereon executable instructions to cause a computer to perform said method for decoding of directions of dominant directional signals. In one embodiment, an apparatus for decoding a compressed HOA signal, comprising a processor and a memory storing instructions that, when executed, cause the apparatus to perform the steps of claim 1.
[0146] It is expressly intended that all combinations of those elements that perform substantially the same function in substantially the same way to achieve the same results are within the scope of the invention, and that each feature disclosed in the description and (where appropriate) the claims and drawings may be provided independently or in any appropriate combination. Features may, where appropriate be implemented in hardware, software, or a combination of the two.
References
[0147]
- [1] Jérôme Daniel. Représentation de champs acoustiques, application à la transmission et à la reproduction de scènes sonores complexes dans un contexte multimédia. PhD thesis, Université Paris 6, 2001.
- [2] Jörg Fliege and Ulrike Maier. A two-stage approach for computing cubature formulae for the sphere. Technical report, Fachbereich Mathematik, Universität Dortmund, 1999. Node numbers are found at http://www.mathematik.unidortmund.de/lsx/research/projects/fliege/nodes/nodes.html.
- [3] Sven Kordon and Alexander Krueger. Adaptive value range control for HOA signals. Patent application (Technicolor Internal Reference: PD130016), July 2013.
- [4] Alexander Krueger and Sven Kordon. Intelligent signal extraction and packing for compression of HOA sound field representations. Patent application EP 13305558.2 (Technicolor Internal Reference: PD130015), filed 29. April 2013.
- [5] A. Krueger, S. Kordon, and J. Boehm. HOA compression by decomposition into directional and ambient components. Published patent application EP2743922 (Technicolor Internal Reference: PD120055), December 2012.
- [6] Alexander Krüger, Sven Kordon, Johannes Boehm, and Jan-Mark Batke. Method and apparatus for compressing and decompressing a higher order ambisonics signal representation. Published patent application EP2665208 (Technicolor Internal Reference: PD120015), May 2012.
- [7] Alexander Krüger. Method and apparatus for robust sound source direction tracking based on Higher Order Ambisonics. Published patent application EP2738962 (Technicolor Internal Reference: PD120049), November 2012.
- [8] Daniel D. Lee and H. Sebastian Seung. Learning the parts of objects by nonnegative matrix factorization. Nature, 401:788-791, 1999.
- [9] ISO/IEC JTC 1/SC 29 N. Text of ISO/IEC 23008-3/CD, MPEG-H 3d audio, April 2014.
- [10] Boaz Rafaely. Plane-wave decomposition of the sound field on a sphere by spherical convolution. J. Acoust. Soc. Am., 4(116):2149-2157, October 2004.
- [11] Earl G. Williams. FourierAcoustics, volume 93 of Applied Mathematical Sciences. Academic Press, 1999.
1. A method for decoding a compressed HOA representation, comprising
- extracting (s41,s42,s43) from the compressed HOA representation a plurality of truncated HOA coefficient sequences (ẑ1(k),...,ẑI(k)), an assignment vector (vAMB,ASSIGN(k)) indicating or containing sequence indices of said truncated HOA coefficient sequences, subband related direction information (MDIR(k+1,f1),..., MDIR(k+1,fF)), a plurality of prediction matrices (A(k+1,f1),...,A(k+1,fF)), and gain control side information (e1(k),β1(k),...,eI(k),βI(k));
- reconstructing (s51,s52) a truncated HOA representation (ĈT(k)) from the plurality of truncated HOA coefficient sequences (ẑ1(k),...,ẑI(k)), the gain control side information (e1(k),β1(k), ..., eI(k),βI(k)) and the assignment vector (vAMB,ASSIGN(k));
- decomposing (s53) in Analysis Filter banks (53) the reconstructed truncated HOA
representation (ĈT(k)) into frequency subband representations
...,
for a plurality of F frequency subbands;
- synthesizing (s54) in Directional Subband Synthesis blocks (54) for each of the
frequency subband representations a predicted directional HOA representation
from the respective frequency subband representation
of the reconstructed truncated HOA representation, the subband related direction
information (MDIR(k+1,f1),...,MDIR(k+1,fF)) and the prediction matrices (A(k+1,f1),...,A(k+1,fF));
- composing (s55) in Subband Composition blocks (55) for each of the F frequency subbands a decoded subband HOA representation
with coefficient sequences
that are either obtained from coefficient sequences of the truncated HOA representation
if the coefficient sequence has an index n that is included in the assignment vector
(vAMB,ASSIGN(k)), or otherwise obtained from coefficient sequences of the predicted directional
HOA component
provided by one of the Directional Subband Synthesis blocks (54); and
- synthesizing (s56) in Synthesis Filter banks (56) the decoded subband HOA representations
to obtain the decoded HOA representation (Ĉ(k)).
2. Method according to claim 1, wherein the extracting comprises demultiplexing (s41)
the compressed HOA representation to obtain a perceptually coded portion and an encoded
side information portion.
3. Method according to claim 1 or 2, wherein the extracting comprises obtaining a perceptually
coded portion that comprises encoded truncated HOA coefficient sequences (
(k), ...,
(k)), and further comprises perceptually decoding (s42) in a perceptual decoder (42) the encoded truncated HOA coefficient sequences (
(k),...,
(k)) to obtain the truncated HOA coefficient sequences (ẑ1(k),...,ẑI(k)).
(k), ...,
(k)), and further comprises perceptually decoding (s42) in a perceptual decoder (42) the encoded truncated HOA coefficient sequences (
(k),...,
(k)) to obtain the truncated HOA coefficient sequences (ẑ1(k),...,ẑI(k)).
4. Method according to one of the claims 1-3, wherein the extracting comprises obtaining
an encoded side information portion, and further comprises decoding (s43) in a side
information source decoder (43) the encoded side information portion to obtain the
subband related direction information (MDIR(k+1,f1),...,MDIR(k+1,fF)), prediction matrices (A(k+1,f1),...,A(k+1,fF)), gain control side information (e1(k),β1(k), ..., eI(k),βI(k)) and assignment vector (vAMB,ASSIGN(k)).
5. Method according to one of the claims 1-4, wherein the subband related direction information
comprises a set of active directions (MDIR(k)) and a tuple set (MDIR(k+1,f1), ...,MDIR(k+1,fF)) that comprises tuples of indices with a first and a second index, the second index
being an index of an active direction within the set of active directions (MDIR(k)) for a current frequency subband, and the first index being a trajectory index
of the active direction, wherein a trajectory is a temporal sequence of directions
of a particular sound source.
6. A method for encoding frames of an input HOA signal having a given number of coefficient
sequences, where each coefficient sequence has an index, comprising
- determining (s111) a set of indices of active coefficient sequences (IC,ACT(k)) to be included in a truncated HOA representation;
- computing (s110) the truncated HOA representation (CT(k)) having a reduced number of non-zero coefficient sequences;
- estimating (s16) from the input HOA signal a first set of candidate directions (MDIR(k));
- dividing (s15) the input HOA signal into a plurality of frequency subbands (f1,...,fF), wherein coefficient sequences
of the frequency subbands are obtained;
- estimating (s161) for each of the frequency subbands a second set of directions (MDIR(k,f1), ..., MDIR(k,fF)), wherein each element of the second set of directions is a tuple of indices with a first and a second index, the second index being an index of an active direction for a current frequency subband and the first index being a trajectory index of the active direction, wherein each active direction is also included in the first set of candidate directions (MDIR(k)) of the input HOA signal;
- for each of the frequency subbands, computing (s17) directional subband signals
(X̃(k - 1,k,f1),..., X̃(k - 1,k,fF)) from the coefficient sequences
...,
of the frequency subband according to the second set of directions (MDIR(k,f1),...,MDIR(k,fF)) of the respective frequency subband;
- for each of the frequency subbands, calculating (s18) a prediction matrix (A(k,f1), ..., A(k,fF)) adapted for predicting the directional subband signals (X̃(k - 1,k,f1), ...,X̃(k - 1,k,fF)) from the coefficient sequences
of the frequency subband using the set of indices of active coefficient sequences
(IC,ACT(k)) of the respective frequency subband; and
- encoding (s19) the first set of candidate directions (MDIR(k)), the second set of directions (MDIR(k,f1),..., MDIR(k,fF)), the prediction matrices (A(k,f1),...,A(k,fF)) and the truncated HOA representation (CT(k)).
7. Method according to claim 6, wherein at least one group of two or more subbands is
created, and wherein the at least one group is used instead of a single subband and
is treated in the same way as a single subband.
8. Method according to claim 6 or 7, wherein said encoding the truncated HOA representation
(CT(k)) comprises
- partial decorrelation (s12) of the truncated HOA channel sequences;
- channel assignment (s13) for assigning the truncated HOA channel sequences (y1(k),..., yI(k)) to transport channels;
- performing gain control (s14) on each of the transport channels, wherein gain control side information (ei(k - 1), βi(k - 1)) for each transport channel is generated;
- encoding (s31) the gain controlled truncated HOA channel sequences (z1(k),..., zI(k)) in a perceptual encoder (31);
- encoding (s32) the gain control side information (ei(k - 1), βi(k - 1)), the first set of candidate directions (MDIR(k)), the second set of directions (MDIR(k,f1),..., MDIR(k,fF)) and the prediction matrices (A(k,f1),...,A(k,fF)) in a side information source coder (32); and
- multiplexing (s33) the outputs of the perceptual encoder (31) and the side information
source coder (32) to obtain an encoded HOA signal frame (
(k - 1)).
9. Method according to one of the claims 6-8, wherein in the step of estimating (s161)
for each of the frequency subbands the second set of directions (MDIR(k,f1),..., MDIR(k,fF)), the directions of a frequency subband are searched only among the directions (MDIR(k)) of the full band HOA signal.
10. Method according to one of the claims 6-9, further comprising a step of determining
a trajectory of an active direction, wherein an active direction is a direction of
a sound source and wherein a trajectory is a temporal sequence of directions of a
particular sound source.
11. Method according to one of the claims 6-10, wherein a truncated HOA representation
is a HOA signal in which one or more coefficient sequences are set to zero.
12. An apparatus (10) for encoding frames of an input HOA signal having a given number
of coefficient sequences, where each coefficient sequence has an index, comprising
- a computation and determining module (11) configured to compute a truncated HOA representation (CT(k)) having a reduced number of non-zero coefficient sequences, and further configured to determine a set of indices of active coefficient sequences (IC,ACT(k)) included in the truncated HOA representation;
- an Analysis Filter bank module (15) configured to divide the input HOA signal into a plurality of frequency subbands (f1,...,fF), wherein coefficient sequences (C̃(k - 1,k,f1),...,C̃(k - 1, k,fF) of the frequency subbands are obtained;
- a Direction Estimation module (16) configured to estimate from the input HOA signal a first set of candidate directions (MDIR(k)), and further configured to estimate for each of the frequency subbands a second set of directions (MDIR(k,f1), ..., MDIR(k,fF)), wherein each element of the second set of directions is a tuple of indices with a first and a second index, the second index being an index of an active direction for a current frequency subband and the first index being a trajectory index of the active direction, wherein each active direction is also included in the first set of candidate directions (MDIR(k)) of the input HOA signal;
- at least one Directional Subband Computation module (17) configured to compute,
for each of the frequency subbands, directional subband signals (X̃(k-1,k,f1),...,X̃(k-1,k,fF)) from the coefficient sequences
of the frequency subband according to the second set of directions (MDIR(k,f1),...,MDIR(k,fF)) of the respective frequency subband;
- at least one Directional Subband Prediction module (18) configured to calculate,
for each of the frequency subbands, a prediction matrix (A(k,f1),..., A(k,fF)) adapted for predicting the directional subband signals (X̃(k - 1,k,f1),..., X̃(k - 1, k, fF)) from the coefficient sequences
of the frequency subband using the set of indices of active coefficient sequences
(IC,ACT(k)) of the respective frequency subband; and
- encoding module (30) configured to encode the first set of candidate directions (MDIR(k)), the second set of directions (MDIR(k,f1),..., MDIR(k,fF)), the prediction matrices (A(k,f1),...,A(k,fF)) and the truncated HOA representation (CT(k)).
13. The apparatus according to claim 12, further comprising
- a partial decorrelator (12) configured to partially decorrelate the truncated HOA channel sequences;
- a Channel Assignment module (13) configured to assigning the truncated HOA channel sequences (y1(k),..., yI(k)) to transport channels; and
- at least one Gain Control unit (14) configured to perform gain control on the transport
channels, wherein gain control side information (ei(k - 1), βi(k - 1)) for each transport channel is generated;
and wherein the encoding module (30) comprises
- a Perceptual Encoder (31) configured to encode the gain controlled truncated HOA channel sequences (z1(k),..., zI(k));
- a Side Information Source Coder (32) configured to encode the gain control side information (ei(k - 1), βi(k - 1)), the first set of candidate directions (MDIR(k)), the second set of directions (MDIR(k,f1),..., MDIR(k,fF)) and the prediction matrices (A(k,f1),...,A(k,fF)); and
- a Multiplexer (33) configured to multiplex the outputs of the perceptual encoder
(31) and the side information source coder (32) to obtain an encoded HOA signal frame
(
(k - 1)).
14. An apparatus (50) for decoding a HOA signal, comprising
- an Extraction module (40) configured to extract from the compressed HOA representation a plurality of truncated HOA coefficient sequences (ẑ1(k),...,ẑI(k)), an assignment vector (vAMB,ASSIGN(k)) indicating or containing sequence indices of said truncated HOA coefficient sequences, subband related direction information (MDIR(k+1,f1),..., MDIR(k+1,fF)), a plurality of prediction matrices (A(k+1,f1),...,A(k+1,fF)), and gain control side information (e1(k),β1(k),..., eI(k),βI(k));
- a Reconstruction module (51,52) configured to reconstruct a truncated HOA representation (ĈT(k)) from the plurality of truncated HOA coefficient sequences (ẑ1(k),...,ẑI(k)), the gain control side information (e1(k),β1(k),...,eI(k),βI(k)) and the assignment vector (vAMS,ASSIGN(k));
- an Analysis Filter bank module (53) configured to decompose the reconstructed truncated
HOA representation (ĈT(k)) into frequency subband representations
for a plurality of F frequency subbands;
- at least one Directional Subband Synthesis module (54) configured to synthesize
for each of the frequency subband representations a predicted directional HOA representation
from the respective frequency subband representation
of the reconstructed truncated HOA representation, the subband related direction
information (MDIR(k+1,f1),..., MDIR(k+1,fF)) and the prediction matrices (A(k+1,f1),...,A(k+1,fF));
- at least one Subband Composition module (55) configured to compose for each of the
F frequency subbands a decoded subband HOA representation
...,
with coefficient sequences
that are either obtained from coefficient sequences of the truncated HOA representation
if the coefficient sequence has an index n that is included in the assignment vector
(vAMB,ASSIGN(k)), or otherwise obtained from coefficient sequences of the predicted directional
HOA component
provided by one of the Directional Subband Synthesis module (54); and
- a Synthesis Filter bank module (56) configured to synthesize the decoded subband
HOA representations
to obtain the decoded HOA representation (Ĉ(k)).
15. Apparatus according to claim 14, wherein the Extraction module (40) comprises at least
- a Demultiplexer (41) for obtaining an encoded side information portion and a perceptually
coded portion that comprises encoded truncated HOA coefficient sequences
(k), ...,
(k);
- a Perceptual Decoder (42) configured to perceptually decode (s42) the encoded truncated
HOA coefficient sequences
(k), ...,
(k)) to obtain the truncated HOA coefficient sequences (ẑ1(k), ...,ẑI(k)); and
- a Side Information Source Decoder (43) configured to decode (s43) the encoded side information portion to obtain the subband related direction information (MDIR(k+1,f1),...,MDIR(k+1,fF)), prediction matrices (A(k+1,f1),...,A(k+1,fF)), gain control side information (e1(k),β1(k), ..., eI(k),βI(k)) and assignment vector (vAMB,ASSIGN(k)).
REFERENCES CITED IN THE DESCRIPTION
This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.
Patent documents cited in the description
Non-patent literature cited in the description
- JÉRÔME DANIELReprésentation de champs acoustiques, application à la transmission et à la reproduction de scènes sonores complexes dans un contexte multimédiaPhD thesis, 2001, [0147]
- JÖRG FLIEGEULRIKE MAIERA two-stage approach for computing cubature formulae for the sphereTechnical report, Fachbereich Mathematik, 1999, [0147]
- SVEN KORDONALEXANDER KRUEGERAdaptive value range control for HOA signalsPatent application, 2013, [0147]
- DANIEL D. LEEH. SEBASTIAN SEUNGLearning the parts of objects by nonnegative matrix factorizationNature, 1999, vol. 401, 788-791 [0147]
- ISO/IEC JTC 1/SC 29 N. Text of ISO/IEC 23008-3/CD, MPEG-H 3d audio, 2014, [0147]
- BOAZ RAFAELYPlane-wave decomposition of the sound field on a sphere by spherical convolutionJ. Acoust. Soc. Am., 2004, vol. 4, 1162149-2157 [0147]
- FourierAcousticsEARL G. WILLIAMSApplied Mathematical SciencesAcademic Press19990000vol. 93, [0147]